More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    YouTube Research

    1

    Evidence-backed YouTube research using search, transcripts, metadata, and public comments.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    YouTube Research

    Example session with this skill installed

    Research YouTube videos about OpenClaw memory systems and long-term memory for AI agents.

    Find relevant videos and choose the 3 strongest candidates for someone who wants practical implementation guidance.

    Compare the 3 videos using the same criteria:

    • relevance to practical implementation
    • technical depth
    • recency
    • usefulness of the advice

    For the video you recommend as the best

    • give a concise summary
    • identify the most important ideas
    • provide timestamps for the key sections
    • analyze the main themes from its public comments
    • distinguish clearly between what the video says and what commenters say

    Finish with your recommended video, a comparison of all 3, key takeaways, timestamped sections worth watching, public-comment themes, and links to all videos used.

    • Read your context and instructions
    • Compiled the youtube research

    Recommended video

    Start with How to Give OpenClaw Permanent Memory with Supabase by Derek Cheung.

    It offers the clearest route to a working prototype: database setup, a Python memory skill, installation, and a demonstration of recall after clearing the conversation context.

    Three topic variations were searched through YouTube MCP, followed by metadata and transcript inspection for six shortlisted videos. The three candidates below prioritize concrete setup and verification over promotional claims. This is a selection from the returned results, not an exhaustive ranking of YouTube.

    Comparison

    VideoPractical relevanceTechnical depthRecencyRecommendation
    Supabase memory tutorialVery high — OpenClaw deployment, SQL schema, skill installation, and context-reset testModerate — pgvector, embeddings, save/recall functionsApr 21, 2026Best first build
    OpenClaw Memory Setup with Mem0Very high — plugin setup, account connection, configuration troubleshootingModerate — automatic capture and recall, with much of installation delegated to the agentJun 3, 2026Best guided plugin setup
    Agent Memory EXPLAINED — Complete ArchitectureHigh for building a memory system; less focused on OpenClaw installationHighest — extraction, storage, metadata, deduplication, retrieval, ranking, and local modelsAug 17, 2026Best engineering companion

    Key takeaways from the recommended video

    The Supabase tutorial separates short-lived conversation context from persistent storage. Memories are stored externally in Supabase Postgres with pgvector and retrieved semantically through embeddings.

    The strongest part of the tutorial is its verification process: the presenter saves preferences and goals, inspects the stored database rows, clears the active conversation context, and then tests whether the agent can recall the previously stored information.

    That demonstrates cross-context persistence, but it does not prove flawless memory, comprehensive capture, or production-grade reliability.

    Timestamped sections worth watching

    • 1:35 — Deploy OpenClaw on Railway
    • 2:07 — Create the Supabase project
    • 3:07 — SQL schema, pgvector, and similarity search
    • 4:24 — Memory skill architecture
    • 6:56 — Inspect saved memories
    • 7:57 — Clear context and test recall

    Public-comment themes

    The retrieved sample contained four unique top-level comment threads, so there is not enough evidence to make audience-wide claims.

    The substantive themes were

    Offline and cloud dependence — one viewer questioned whether the design makes sense for an offline agent and suggested local/cloud replication.

    • Cost under sustained usage — another viewer asked whether free tiers could support heavier daily workloads.
    • Implementation resources — the creator shared links to the skill repository and deployment materials.
    • General appreciation — one viewer left brief positive feedback.

    The video itself demonstrates saving and recalling memories. The commenters mainly raise practical questions around cost and offline operation; they do not independently validate the system’s reliability.

    Evidence limitations

    Complete transcripts were read for all three selected videos. The Supabase and Mem0 transcripts were auto-generated, while the Hugging Face transcript was marked non-automatic.

    Linked code and deployment resources were not executed or independently audited. Pricing, service limits, and compatibility claims mentioned inside the videos were not independently verified.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Sourcing useful information from public YouTube videos often results in surface-level summaries or missed context. YouTube Research uses youtube-mcp to perform structured, evidence-backed research from search results, public metadata, transcripts, and comments.

    It goes beyond basic video descriptions to find specific timestamps, compare multiple videos using consistent criteria, analyze long transcripts, and identify recurring themes in retrieved public comments. Material conclusions are tied to retrieved evidence or clearly labeled as interpretation.

    What it does

    • Video research & verification retrieves public metadata such as views, publication date, duration, tags, channel information, and description to provide context for evaluation.
    • Transcript analysis examines video content for claims, key ideas, quotations, and timestamped sections.
    • Public-comment analysis identifies recurring discussion themes, questions, and reactions while treating retrieved comments as a sample rather than the entire audience.
    • Comparative research evaluates multiple videos against the same criteria so recommendations are consistent and transparent.
    • Targeted discovery searches YouTube and shortlists candidates based on the user’s requirements.
    • Language-aware research checks available caption languages when a requested transcript track is unavailable or uncertain.

    How it works

    1. Identify the evidence needed — metadata, transcript, comments, search results, or a combination.
    2. Collect the minimum sufficient evidence using targeted youtube-mcp tools and transcript/comment pagination when necessary.
    3. Analyze consistently by separating what the video says, what public commenters say, and what the agent concludes.
    4. Generate a useful research artifact with video links, comparison criteria, timestamps where relevant, and clearly stated evidence limitations.

    Frameworks & tools

    This skill uses the open-source @granitebps/youtube-mcp server and can work with:

    • search_youtube
    • get_video_info
    • get_video_transcript
    • get_transcript_languages
    • get_video_comments
    • get_server_info

    No YouTube API key is required with the default provider. Users may optionally configure their own YouTube Data API key for metadata, comments, and search.

    Why this beats prompting it yourself

    A generic prompt can ask an agent to research YouTube, but it does not define which evidence to collect, when enough evidence has been gathered, how to compare videos consistently, how to paginate long transcripts, or how to avoid overgeneralizing from a small comment sample.

    This skill provides that repeatable workflow and uses retrieved YouTube MCP data as the evidence source.

    Use cases

    • Compare several product reviews or tutorials and recommend the best one for a specific goal.
    • Locate the exact timestamp where a speaker discusses a particular topic.
    • Summarize a long technical keynote, interview, podcast, or tutorial with key takeaways and timestamps.
    • Research the strongest videos on a topic and produce a concise research brief.
    • Analyze retrieved public comments to surface recurring questions, pain points, and discussion themes.

    Known limitations

    • Cannot access private YouTube Analytics such as revenue, retention, CTR, traffic sources, or audience demographics.
    • Search results represent the returned result set and are not an exhaustive ranking of all YouTube.
    • Public comments are a retrieved sample and should not be treated as representative of the entire audience.
    • Transcript analysis depends on available captions; auto-generated captions may contain recognition errors.
    • Transcript retrieval uses the MCP’s transcript backend even when the optional official YouTube Data API provider is selected.

    No access to private YouTube Analytics such as CTR, revenue, retention, traffic sources, or audience demographics.

    Results depend on the public data available for each video, including metadata, captions/transcripts, and comments. Some videos may have no captions or comments.

    Search results are limited to the result set returned by the configured youtube-mcp provider and should not be treated as an exhaustive ranking of all YouTube.

    Public-comment analysis is based on the retrieved sample and should not be treated as representative of the entire audience.

    Auto-generated captions may contain transcription errors. YouTube may also restrict or block some access paths, especially from certain cloud or datacenter IPs.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 28 days ago

    Requires

    @granitebps/youtube-mcp v1.1.1+
    Node.js 22+
    Capable Agent MCP
    Youtube MCP

    Permissions

    Read Files
    Network Access

    The skill reads its bundled reference files and uses @granitebps/youtube-mcp to retrieve public YouTube search results, metadata, transcripts, and comments. It does not write or modify user files, does not require browser access, and does not require a connected YouTube or Google account. The default youtube-mcp provider requires no API key. Users may optionally configure their own YOUTUBE_API_KEY in the MCP server environment to use the official YouTube Data API provider for metadata, comments, and search. The skill itself does not read or store credentials.

    Listed28 days ago

    What's inside

    Frequently Asked Questions

    A normal prompt can ask an AI to summarize a video, but it does not define how to gather and verify the evidence. YouTube Research uses `youtube-mcp` to retrieve public metadata, transcripts, search results, and comments, then follows a structured workflow for comparison, timestamped analysis, and evidence-backed conclusions. It also keeps video claims, public comments, and the agent’s own interpretation clearly separated.

    Use it with an AI agent or environment that supports Agent Skills/SKILL.md and MCP servers. The skill has been end-to-end tested with Codex. Other compatible agents should also work when `youtube-mcp` is configured in the host environment.

    You receive the complete YouTube Research skill package, including the main `SKILL.md` workflow and reference guides for research workflows, output formats, `youtube-mcp` setup, and evidence/access limitations. The open-source `youtube-mcp` runtime is installed separately.

    The skill requires an MCP-capable agent, Node.js 22+, and `@granitebps/youtube-mcp` v1.1.1 or newer within the v1 compatibility line. No YouTube API key is required with the default provider. Users may optionally configure their own `YOUTUBE_API_KEY` with `YT_PROVIDER=api` to use the official YouTube Data API for metadata, comments, and search. Transcript retrieval uses the MCP transcript backend regardless of provider.

    No. The skill works with public YouTube data available through `youtube-mcp`. It cannot access private YouTube Analytics such as revenue, CTR, retention, traffic sources, subscriber conversion, or audience demographics.

    Yes. Multi-video comparison is one of the main workflows. The skill defines the comparison criteria first, gathers consistent evidence for each video, and evaluates them using the same criteria. It can then recommend the strongest option for the user’s stated goal while showing missing or incomplete evidence.

    New in Research & Analysis